Visualization 2: Spectral Data Explorer

Understanding Spectral Data

Near-infrared (NIR) spectroscopy is a non-destructive technique that measures how plant tissues absorb and reflect light at different wavelengths. Each crop has a unique "spectral fingerprint" determined by its biochemical composition.

Wavelength Range

3,999 - 10,001 nm
Mid-infrared region sensitive to organic compounds

Total Measurements

1,557 wavelengths
High-resolution spectral data per sample

Interactive Spectral Visualization

Instructions:

Use the Mean Spectrum / All Samples buttons to switch views
Click on crop names in the legend to show/hide specific crops
Hover over the chart to see exact wavelength and value
Use the toolbar to zoom, pan, or download the figure

Spectral Region Interpretation

The near-infrared (NIR) spectrum reveals chemical composition through molecular absorption:

4,000-5,000 nm: C-H stretching - lipids, carbohydrates
5,000-6,000 nm: C=O stretching - proteins, organic acids
6,000-7,500 nm: N-H bending - proteins, amino acids (nitrogen indicator)
7,500-10,000 nm: C-O stretching - carbohydrates, starch

Differences in spectral signatures between crops reflect their unique biochemical composition.

Dataset Summary

Samples per Crop Type

Citrus: 2,346 samples
Almond: 1,260 samples
Avocado: 508 samples
Vine: 238 samples

Key Spectral Observations

Distinct Signatures: Each crop shows a unique spectral pattern, reflecting differences in leaf structure and chemical composition
Common Absorption Bands: All crops show absorption features at similar wavelengths (water, chlorophyll, cellulose), but with varying intensities
Nitrogen Sensitivity: The 6,000-7,500 nm region is particularly sensitive to nitrogen-containing compounds, making it valuable for N status prediction
Carbohydrate Features: Starch and sugar content influence the 7,500-10,000 nm region, relevant for our N/ST ratio analysis

From Spectra to Predictions

This spectral data forms the foundation of our machine learning models:

PLSR (Partial Least Squares Regression): Reduces 1,557 wavelengths to key components
Random Forest & XGBoost: Predict N, SC, and ST values from spectral features
N/ST Ratio Prediction: Enables fertilization timing recommendations without destructive sampling

The ability to predict chemical composition from spectral data enables rapid, non-destructive assessment of crop nitrogen status in the field.